[05:24:58] <cscott>	 apergos: (digging up the thread from earlier) the issue is that there's a legit use for "daily" expiration, for things like article of the day, etc.  But we get in trouble when 20% of our pages expire at midnight (which is where we're at right now).
[05:25:27] <cscott>	 re: parsing -- there are also algorithmic fixes to speed up parsing, which hopefully CTT will be working on next FY (selective update)
[05:25:29] <apergos>	 24 hours is ok. less than  that? mmmmmmm
[05:25:43] <apergos>	 I'm definitely interested in hearing about those
[21:12:10] <Krinkle>	 most cases that expire daily don't need to. e.g. age doesn't change every day. likewise, a citation input validator that asserts `access_date <= today` never needs to run again. Today, with my edits in place, both of those were fixed from expiring after 30min, to expiring after 24h / end of day, by making sure they only acess the current day, not time.
[21:13:30] <Krinkle>	 but cscot.t  and or.i's improvement could push that back to ~30 days for most pages. However that require adoption of a new mechanism, and will only work if there aren't still other templates on the same page that do the simple thing instead of adopting a new mechanism, so it's not a given that it will make a big picture difference in practice, but is worth trying.
[21:14:33] <Krinkle>	 basically instead of giving toda's date to Lua, and compare in user code, the user code gives the date it compares against (past access_date, or next birthday) so that it can do the comparison for you and thus lower TTL to when that changes instead of needing to cap to 24h because you used the day's date somehow.